Dear PLS experts,
I have both working memory and episodic memory block designs from the same participants. I am interested to see which brain networks are commonly and/or uniquely engaged during the tasks, and also how such networks are correlated with performance and age. I guess the multi-block pls should be an optimal approach to address both questions simultaneously, right? If so, I have a couple of questions:
LV1: Task PLS part of the multiblock (Brain score): Con A Con B Con C: -38 -5 43
It is clear that this LV reflects those brain areas with greater activation during condition A&B (task) than condition C (baseline) and vice versa. Note that the sign for brain scores of the task conditions (A&B) are negative.
LV1:behavioral PLS part of multiblock:
Correlation between Brain scores of condition A and behavioral measure 1: r = - 0.38
Correlation between Brain scores of condition A and behavioral measure 2: r = 0.25
Given the sign of the voxel saliencies and also brain score for task conditions (A&B conditions) derived from task PLS are negative; I am just wondering whether I need to reverse the sign of the correlation between brain score of the task conditions and behavioral measure. On the other word, do you say that the voxels that are more activated during task (A&B) than baseline C (i.e. these regions have negative saliencies) are positively or negatively correlated with behavioral measure 1?
Thanks a lot for your help.
Best
Kami
Hi Kami - see my responses below
No, PLS is not mathematically affected by collinearity - in fact its a benefit so long as the variables are not perfectly correlated.
Probably using age as a continuous variable would be better. The bias in sample size can skew things for the resampling stats
...
Given the sign of the voxel saliencies and also brain score for task conditions (A&B conditions) derived from task PLS are negative; I am just wondering whether I need to reverse the sign of the correlation between brain score of the task conditions and behavioral measure. On the other word, do you say that the voxels that are more activated during task (A&B) than baseline C (i.e. these regions have negative saliencies) are positively or negatively correlated with behavioral measure 1?
The signs in the SVD decomposition are arbitrary so it perfectly legitimate to reverse the signs so long as you do it for all elements within a LV. In your case it will help with the interpretation.
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2024 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED